Journal of Experimental Psychology: Learning, Memory, and Cognition

Journal of Experimental Psychology: Learning, Memory, and Cognition Additive and Interactive Effects in Semantic Priming: Isolating Lexical and Decision Processes in the Lexical Decision Task Melvin J. Yap, David A. Balota, and Sarah E. Tan Online First Publication, May 21, 2012. doi: 10.1037/a0028520

CITATION Yap, M. J., Balota, D. A., & Tan, S. E. (2012, May 21). Additive and Interactive Effects in Semantic Priming: Isolating Lexical and Decision Processes in the Lexical Decision Task. Journal of Experimental Psychology: Learning, Memory, and Cognition. Advance online publication. doi: 10.1037/a0028520

Journal of Experimental Psychology: Learning, Memory, and Cognition 2012, Vol. ●●, No. ●, 000 – 000

© 2012 American Psychological Association 0278-7393/12/$12.00 DOI: 10.1037/a0028520

Additive and Interactive Effects in Semantic Priming: Isolating Lexical and Decision Processes in the Lexical Decision Task Melvin J. Yap

David A. Balota

National University of Singapore

Washington University in St. Louis

Sarah E. Tan National University of Singapore The present study sheds light on the interplay between lexical and decision processes in the lexical decision task by exploring the effects of lexical decision difficulty on semantic priming effects. In 2 experiments, we increased lexical decision difficulty by either using transposed letter wordlike nonword distracters (e.g., JUGDE; Experiment 1) or by visually degrading targets (Experiment 2). Although target latencies were considerably slowed by both difficulty manipulations, stimulus quality— but not nonword type—moderated priming effects, consistent with recent work by Lupker and Pexman (2010). To characterize these results in a more fine-grained manner, data were also analyzed at the level of response time (RT) distributions, using a combination of ex-Gaussian, quantile, and diffusion model analyses. The results indicate that for clear targets, priming was reflected by distributional shifting of comparable magnitude across different nonword types. In contrast, priming of degraded targets was reflected by shifting and an increase in the tail of the distribution. We discuss how these findings, along with others, can be accommodated by an embellished multistage activation model that incorporates retrospective prime retrieval and decision-based mechanisms. Keywords: visual word recognition, semantic priming, response time distributional analyses, ex-Gaussian analysis, diffusion model

(Norris, 2006, 2009) have attempted to unify word recognition and decision-making processes within an integrated framework that likens readers to optimal Bayesian decision makers. In the present study, we investigate the roles of lexical and decision processes in the lexical decision task by examining the joint effects of semantic priming with two variables known to strongly modulate lexical decision performance: stimulus quality and nonword type. The semantic priming effect is the well-known finding that words preceded by related primes (e.g., NURSE– DOCTOR) are recognized faster than those preceded by unrelated primes (e.g., BUTTER–DOCTOR; Meyer, Schvaneveldt, & Ruddy, 1975). Stimulus quality refers to whether items are presented clearly or in a degraded manner; clear words are responded to faster and more accurately than degraded words (Becker & Killion, 1977). Nonword type refers to the similarity of the nonword distracters to the word targets, and the general finding is that lexical decision responses to words are slower and less accurate when more wordlike distracters are used (Stone & Van Orden, 1993).

Visual word recognition has been the focus of considerable research, with the lexical decision task being the most commonly used task to study visual word recognition (Murray & Forster, 2004), featured in thousands of studies (Gomez, in press). Importantly, lexical decision performance has been central to influential models that focus on lexical processing as well as models that focus on decision mechanisms. The former include the multistage activation model (Borowsky & Besner, 1993) and the multiple read-out model (MROM; Jacobs & Grainger, 1994), whereas the latter include the two-stage model (Balota & Spieler, 1999), the random walk model (Stone & Van Orden, 1993), the diffusion model (Ratcliff, Gomez, & McKoon, 2004), and the leaky competing accumulator model (Dufau, Grainger, & Ziegler, 2012). In addition, more recent models such as the Bayesian reader model

Melvin J. Yap, Department of Psychology, National University of Singapore, Singapore; David A. Balota, Department of Psychology, Washington University in St. Louis; Sarah E. Tan, Department of Psychology, National University of Singapore. Portions of this research were presented as a poster at the annual conference of the British Psychological Society in May 2011. We thank Yan Ling Leow for her assistance with data collection. We also thank Kathy Rastle, Dennis Norris, and Derek Besner for their helpful comments on an earlier version of this article. Correspondence concerning this article should be addressed to Melvin J. Yap, Department of Psychology, National University of Singapore, Block AS4, #02-07, Singapore 117570. E-mail: [email protected]

Semantic Priming, Stimulus Quality, and Nonword Type Several studies (Balota, Yap, Cortese, & Watson, 2008; Borowsky & Besner, 1993; Brown & Besner, 2002; Meyer et al., 1975; Thomas, Neely, & O’Connor, 2012) have observed larger semantic priming effects for degraded targets than for clear targets. According to Sternberg’s (1969; Roberts & Sternberg, 1993) ad1

2

YAP, BALOTA, AND TAN

ditive factors logic, two factors that produce a statistical interaction exert their influence on at least one common processing stage. In contrast, if two factors yield additive effects (i.e., two main effects without an interaction), they influence different stages (although see McClelland, 1979, for an alternative explanation of additive effects). The overadditive Stimulus Quality ⫻ Priming interaction is thus consistent with the idea that semantic priming and stimulus quality influence at least one common processing stage. Interestingly, although semantic priming interacts with stimulus quality, word-frequency and stimulus quality produce clear additive effects in lexical decision, suggesting that these two factors influence separate stages (Yap & Balota, 2007). Given that word-frequency also interacts with semantic priming (see Becker, 1979), this collective set of findings has represented an important conundrum for any comprehensive model of lexical decision performance to accommodate. The interaction between stimulus quality and priming suggests that other experimental manipulations (e.g., stimulus degradation) that make it more difficult to discriminate between words and nonwords should also magnify priming effects. As described earlier, another way to vary lexical decision difficulty is to manipulate the similarity of nonword distracters to real words. Nonwords can be orthographically illegal (e.g., BRNAE), orthographically legal (e.g., BRONE), or homophonous with real words (i.e., pseudohomophones; e.g., BRANE). As nonwords become increasingly wordlike, lexical decision latencies become slower. More importantly, nonword type strongly interacts with other variables that influence lexical decision performance. For example, wordfrequency effects are much larger in the presence of pseudohomophones, compared to legal or illegal nonwords (Stone & Van Orden, 1993; Yap, Balota, Cortese, & Watson, 2006). Interestingly, Lupker and Pexman (2010) have recently demonstrated that increasing lexical decision difficulty by using wordlike nonwords (e.g., BRANE), compared to less wordlike nonwords (e.g., BRONE), did not alter the size of the semantic priming effect (see also Yap, Tse, & Balota, 2009). The additive effects of priming and nonword type are puzzling, in light of the Stimulus Quality ⫻ Priming, Nonword Type ⫻ Word-Frequency, and the Word-Frequency ⫻ Priming interactions. Specifically, if using more wordlike distracters increases lexical decision difficulty, and increased lexical decision difficulty is reflected in greater priming (as reflected by the Stimulus Quality ⫻ Semantic Priming interaction), then it is surprising that priming effects are not moderated by nonword type. Indeed, Lupker and Pexman have pointed out that the additive effects of nonword type and priming constitute another major challenge to any model that attempts to explain nonword type and semantic priming effects in lexical decision using a common lexical mechanism. To our knowledge, the counterintuitive dissociation between the Priming ⫻ Stimulus Quality interaction and the additive effects of priming and nonword type has not been systematically investigated within a single study. More importantly, no extant model can handle these findings. For example, although the multistage interactive activation model (Borowsky & Besner, 1993) is able to accommodate the complex joint effects of stimulus quality, priming, and word-frequency, it is silent on nonword type effects. In contrast, lexical decision models that can accommodate the joint effects of nonword type with other variables—such as Balota and Chumbley’s (1984) two-stage model, Grainger and Jacobs’s

(1996) MROM, Ratcliff et al.’s (2004) diffusion model, and Norris’s (2006, 2009) Bayesian reader model— have not yet explicitly addressed the joint effects of nonword type with either stimulus quality or priming.

Objectives of the Present Study In the present study, we reexamined the joint effects of priming with stimulus quality and nonword type. Although different groups have already reported that priming and stimulus quality interact (e.g., Balota et al., 2008) and that priming and nonword type are additive (Lupker & Pexman, 2010), comparing studies that employ different sets of stimuli or participants with different levels of vocabulary knowledge (i.e., knowledge of word forms and meanings) could be potentially misleading. For example, although the overadditive interaction between priming and word-frequency (i.e., stronger priming for low-frequency words) is considered a benchmark finding in the word recognition literature (see McNamara, 2005; Neely, 1991), Yap et al. (2009) recently demonstrated that the interaction between priming and word-frequency was observed only for participants with less vocabulary knowledge, whereas participants with more vocabulary knowledge produced additive effects. To ensure that any observed effects in the present study are not driven by differences in stimuli difficulty or participants’ vocabulary knowledge, the same prime-target pairs were used across experiments, and participants in the different experiments were matched on vocabulary knowledge (as reflected by performance on a standardized vocabulary measure). More importantly, the present study also afforded the opportunity to explore the effects of the targeted variables on both mean response times (RTs) and the underlying characteristics of RT distributions. Although chronometric studies of cognitive processing overwhelmingly examine RTs at the level of the mean, there is mounting evidence that experimental variables can selectively modulate different characteristics of empirical RT distributions (see Balota & Yap, 2011, for a review). A popular way to carry out distributional analyses is to fit empirical RT data to a theoretical distribution like the ex-Gaussian function, which is the convolution of a Gaussian (normal) and exponential distribution (Ratcliff, 1979). The ex-Gaussian distribution contains three parameters; ␮ and ␴, respectively, reflect the mean and standard deviation of the Gaussian distribution, whereas ␶ reflects the mean and standard deviation of the exponential distribution. Changes in ␮ indicate distributional shifting, whereas changes in ␶ reflect modulations in the tail of the distribution. Ex-Gaussian analysis is usually supplemented by a nonparametric technique called Vincentizing (Ratcliff, 1979), through which one can examine the effect of a variable on different regions of the RT distribution. To carry out Vincentizing, one first rank orders each participant’s RTs separately for each condition, followed by computing effects at different quantiles (e.g., .1, .2, .3, .4, etc.). These tools collectively allow researchers to ascertain whether effects are mediated by distributional shifting, an increase in the slow tail of the distribution, or some combination of both. Importantly, RT distributional analyses have yielded new insights into the nature of semantic priming. For example, Balota et al. (2008) demonstrated that when targets are presented clearly, the semantic priming effect for highly skilled lexical processors is reflected by a shift in the RT distribution (see Figure 1). However,

ADDITIVE AND INTERACTIVE EFFECTS

Figure 1. Interactive effects of Stimulus Quality and Relatedness across quantiles, showing that the larger effects of priming for degraded, compared to clear, targets are due to both a shift and a stretching in the tail of the distribution.

when the same targets are visually degraded, priming simultaneously shifts and increases the tail of the distribution, and this is reflected in more priming for the slower, more difficult trials (see Figure 1). Balota et al. have argued that these results are consistent with the idea that for highly fluent lexical processors, semantic priming for clearly presented words is mediated by a relatively modular head-start mechanism, where primes simply speed up recognition of the related target by some constant amount of time, regardless of target difficulty (for a similar effect on eye fixation durations, see Staub, 2011). In contrast, when targets are degraded, the system adaptively recruits any information available to help resolve the target (Whittlesea & Jacoby, 1990) and is therefore more reliant on prime information for the more difficult items, which is reflected in the slower quantiles of the RT distribution. To our knowledge, how nonword type moderates priming at the level of RT distributions has not been investigated. The study by Lupker and Pexman (2010), which has addressed this question most directly among extant studies (e.g., Joordens & Becker, 1997; Milota, Widau, McMickell, Juola, & Simpson, 1997; Stone & Van Orden, 1992), was based on mean RT analyses. Although ex-Gaussian parameters help provide a finer-grained summary of empirical RT distributions, it is inappropriate, without a theoretical framework, to directly map them onto cognitive processes (Matzke & Wagenmakers, 2009). To better understand the underlying mechanisms, researchers (e.g., Balota & Spieler, 1999; Schmiedek, Oberauer, Wilhelm, Su¨␤, & Wittmann, 2007) have recommended fitting RT data to process-oriented models such as the diffusion model of binary decision (Ratcliff, 1978; Ratcliff et al., 2004). According to the diffusion model, lexical decision involves the accumulation of noisy information over time from a starting point toward one of two decision boundaries (word or nonword). Three major parameters describe this process. Drift rate (v) refers to the mean rate at which information is accumulated from the word or nonword stimulus, boundary separation (a) refers to the location of the response criterion, and Ter refers to the nondecision component that collectively indexes encoding and response execution processes. Importantly, this model allows one to distinguish between the quality of evidence driving the decision,

3

the decision criteria, and nondecision processes (Ratcliff, Thapar, & McKoon, 2010). In order to capitalize on the additional insights afforded by diffusion model parameter estimates, distributional effects in the present study are explored using a combination of ex-Gaussian analysis, Vincentizing, and diffusion model analysis. To recapitulate, the present study uses the same participant pool and stimulus set to directly compare two manipulations of lexical decision difficulty (i.e., nonword type vs. stimulus quality) on semantic priming, and by analyzing these effects both at the level of the mean and at the level of underlying RT distributional characteristics. Specifically, in Experiment 1, we compare priming effects when legal versus transposed-letter (TL) nonwords (e.g., TRIAN) are used. It is important to replicate the intriguing additive pattern observed by Lupker and Pexman (2010), given some of the mixed findings previously reported in this literature (e.g., Milota et al., 1997; Shulman & Davison, 1977). TL nonwords are created by transposing two letters in a word (e.g., TRAIN–TRIAN) and are perceptually very similar to real words (Forster, Davis, Schoknecht, & Carter, 1987; Perea & Lupker, 2003). Crucially, TL nonwords slow participants down even more effectively than pseudohomophones (Lupker & Pexman, 2010; see also Perea & Lupker, 2004). In Experiment 2, we compare priming effects when clear versus degraded targets are presented. Stimuli are degraded by rapidly alternating the target letter string (e.g., DOG) with a randomly generated mask of the same length (e.g., &?#). This paradigm yields robust degradation effects (e.g., Balota et al., 2008; Yap & Balota, 2007; Yap, Balota, Tse, & Besner, 2008) that are qualitatively similar to those produced by other degradation manipulations (e.g., contrast reduction; O’Malley, Reynolds, & Besner, 2007). A within-participant manipulation of stimulus quality is used, consistent with other studies manipulating this variable (Balota et al., 2008; Becker & Killion, 1977; Besner & Smith, 1992; Borowsky & Besner, 1993; Brown, Stolz, & Besner, 2006; O’Malley et al., 2007; Plourde & Besner, 1997).

Experiment 1 Method Participants. Eighty participants (58 females) from the National University of Singapore participated for course credit. All participants were proficient in English and had normal or corrected-to-normal vision. The mean vocabulary age, as measured by the vocabulary subscale of the Shipley Institute of Living Scale (Shipley, 1940), was 18.08 (SD ⫽ 0.82). Design. Priming (related or unrelated) was manipulated within-participants, and Nonword Type (legal nonwords or TL nonwords) was manipulated between-participants; half the participants were presented with legal nonwords, whereas the other half were presented with TL nonwords. Stimuli. One hundred twenty words (see the Appendix for a full list of stimuli) served as targets (see Table 1 for descriptive statistics). Each target word was preceded by either a related or unrelated prime, yielding 60 observations per participant cell. Across participants, stimuli were counterbalanced across the related and unrelated conditions; unrelated primeword target pairs were created by re-pairing the primes and targets within each set. No prime or target was repeated within


4

Table 1 Descriptive Statistics for the Word and Nonword Stimuli Used in Experiments 1 and 2 Word stimuli (N ⫽ 120)

M

SD

Number of morphemes Number of syllables Number of letters Number of phonemes Log HAL frequency (Lund & Burgess, 1996) Number of orthographic neighbors Number of phonological neighbors Orthographic Levenshtein distance (Yarkoni et al., 2008) Phonological Levenshtein distance (Yap & Balota, 2009) Forward associative strength (Nelson et al., 2004) Backward associative strength (Nelson et al., 2004)

1.34 1.79 6.14 5.10

0.51 0.73 1.67 1.70

8.07 2.64 5.75

1.13 3.91 8.15

2.17

0.71

2.03

0.83

0.21

0.18

0.16

0.21

Legal nonwords

Nonword stimuli (N ⫽ 120) Log HAL baseword frequency (Lund & Burgess, 1996) Number of syllables Number of letters Number of orthographic neighbors

TL nonwords

1.76 6.16

0.73 1.65

8.10 1.76 6.14

1.37 0.69 1.67

2.67

3.95

2.73

3.8

Note. HAL ⫽ Hyperspace Analogue to Language; TL ⫽ transposedletter.

a participant. Legal nonwords and the base words from which TL nonwords were derived were matched to target words on number of syllables, letters, and orthographic neighbors (see Table 1). One hundred and twenty legal nonwords were obtained from the English Lexicon Project (Balota et al., 2007), whereas 120 TL nonwords were obtained by rearranging two adjacent letters in words. Four TL nonwords were obtained by transposing the first and second letters, 23 by transposing the second and third letters, 47 by transposing the third and fourth letters, 31 by transposing the fourth and fifth letters, 10 by transposing the fifth and sixth letters, 4 by transposing the sixth

and seventh letters, and 1 by transposing the seventh and eighth letters. Procedure. PC-compatible computers running E-prime software (Schneider, Eschman, & Zuccolotto, 2001) were used for stimulus presentation and data collection. All stimuli were displayed in the center of the computer screen, and participants’ responses were made on a computer keyboard. Participants were tested individually in sound-attenuated cubicles, sitting about 60 cm from the screen. They first provided demographic information (gender, race, age, and years of university education) and completed the vocabulary subscale (40 item vocabulary test) of the Shipley Institute of Living Scale (Shipley, 1940; Zachary, 1992) on the computer. Participants were instructed that a word and a letter string would be presented sequentially and they were to decide whether the letter string formed a word or nonword by making the appropriate button press, that is, the apostrophe key for words and the A key for nonwords. Participants were encouraged to respond quickly but not at the expense of accuracy. There were 20 practice trials, followed by four experimental blocks of 60 trials each, with breaks between blocks. The order in which stimuli were presented was randomized anew for each participant. Stimuli were presented in uppercase 14-point Courier New, and each trial comprised the following order of events: (a) a fixation point (⫹) at the center of the monitor for 2,000 ms, (b) the prime for 150 ms, (c) a blank screen for 650 ms, and (d) the target, resulting in an 800-ms prime-target stimulus onset asynchrony. The target remained on the screen for 3,000 ms or until a response was made. Each correct response was followed by an intertrial interval of 450 ms. If a response was incorrect, a 170-ms tone was presented simultaneously with the word “Incorrect” displayed slightly below the fixation point for 450 ms.

Results and Discussion Errors (5.7% across all conditions) were first excluded from the analyses. In addition, responses below 200 ms and above 3,000 ms were eliminated before a standard deviation was calculated for each participant. Response latencies beyond 2.5 SDs from each participant’s mean were also excluded, and this removed a further 2.8% of the responses. The mean RTs, accuracy rates, and exGaussian parameters are presented in Table 2.

Table 2 Mean RTs, Accuracy, Ex-Gaussian, and Diffusion Model Parameters as a Function of Nonword Type and Relatedness in Experiment 1 Nonword type/priming Legal nonwords Related Unrelated Priming effect TL nonwords Related Unrelated Priming effect Interaction

M

Accuracy

␮

␴

␶

v

a

Ter

606 625 19

0.98 0.96 0.02

479 499 20

52 46 ⫺6

128 127 ⫺1

3.32 3.30 0.02

1.49 1.47 ⫺0.02

0.39 0.41 0.02

764 782 18 ⫺1

0.96 0.96 0 ⫺0.02

531 545 14 ⫺6

62 53 ⫺9 ⫺3

235 238 3 4

2.58 2.52 0.06 0.04

1.88 1.86 ⫺0.02 0

0.40 0.43 0.03 0.01

Note. RT ⫽ response time; TL ⫽ transposed-letter.


Response latencies. For RTs, the main effects of nonword type, Fp(1, 78) ⫽ 25.62, p ⬍ .001, MSE ⫽ 38,866, ␩2p ⫽ .25; Fi(1, 119) ⫽ 368.96, p ⬍ .001, MSE ⫽ 8,318, ␩2p ⫽ .76, and priming, Fp(1, 78) ⫽ 12.75, p ⫽ .001, MSE ⫽ 1,038, ␩2p ⫽ .14; Fi(1, 119) ⫽ 12.39, p ⫽ .001, MSE ⫽ 3,784, ␩2p ⫽ .09, were significant. The Nonword Type ⫻ Priming interaction did not approach significance by participants or by items, Fs ⬍ 1. Accuracy rates. For accuracy rates, the main effect of nonword type was significant by items and approached significance by participants, Fp(1, 78) ⫽ 2.81, p ⫽ .098, MSE ⫽ 0.002, ␩2p ⫽ .035; Fi(1, 119) ⫽ 5.52, p ⫽ .020, MSE ⫽ 0.002, ␩2p ⫽ .044. The main effect of priming, Fp(1, 78) ⫽ 5.67, p ⫽ .020, MSE ⫽ 0.001, ␩2p ⫽ .068; Fi(1, 119) ⫽ 5.19, p ⫽ .025, MSE ⫽ 0.002, ␩2p ⫽ .042, and the Nonword Type ⫻ Priming interaction, Fp(1, 78) ⫽ 6.30, p ⫽ .014, MSE ⫽ 0.001, ␩2p ⫽ .075; Fi(1, 119) ⫽ 4.69, p ⫽ .032, MSE ⫽ 0.002, ␩2p ⫽ .038, were also significant. As shown in Table 2, the priming effect was larger in the context of legal nonwords compared to TL nonwords. Ex-Gaussian analyses. Using the quantile maximum likelihood estimation (QMLE) procedure in the QMPE v2.18 program (Cousineau, Brown, & Heathcote, 2004; Heathcote, Brown, & Mewhort, 2002), ex-Gaussian parameters (␮, ␴, ␶) were obtained for each participant across the different experimental conditions. QMLE provides unbiased parameter estimates and has been demonstrated to be more effective than continuous maximum likelihood estimation for small samples (Heathcote & Brown, 2004; Speckman & Rouder, 2004). All fits successfully converged within 250 iterations. For ␮, the main effects of nonword type, F(1, 78) ⫽ 8.37, p ⫽ .005, MSE ⫽ 11,563, ␩2p ⫽ .10, and priming, F(1, 78) ⫽ 22.02, p ⬍ .001, MSE ⫽ 539, ␩2p ⫽ .22, were significant. The Nonword Type ⫻ Priming interaction was not significant, F ⬍ 1. Turning to ␴, only the main effect of priming was significant, F(1, 78) ⫽ 5.68, p ⫽ .020, MSE ⫽ 424, ␩2p ⫽ .07. Finally, for ␶, only the main effect of nonword type was significant, F(1, 78) ⫽ 33.67, p ⬍ .001, MSE ⫽ 13,962, ␩2p ⫽ .30. Quantile analyses (Vincentizing). The mean quantiles for the different experimental conditions are plotted in Figure 2, whereas Figure 3 presents priming effects as a function of nonword type. In Figure 2, the empirical quantiles are represented by data points and standard error bars, and the estimated quantiles for the best-fitting ex-Gaussian distribution are represented by lines. The theoretical quantiles were estimated for each condition by using Monte Carlo simulations to generate ex-Gaussian distributions (comprising 20,000 observations for each condition) corresponding to the ex-Gaussian parameters for that condition (see White & Staub, 2011). The goodness of fit between the empirical and theoretical quantiles reflects the extent to which the empirical RT distributions are being captured by the ex-Gaussian parameters. From Figure 3, it is clear that priming effects in the legal and TL nonword conditions are approximately the same size and remain relatively invariant across the quantiles. Diffusion model analyses. To obtain diffusion model parameter estimates, we used the e Fast-dm program (Voss & Voss, 2007), which uses the partial differential equation (PDE) method (Voss & Voss, 2008) to estimate diffusion model parameters efficiently and accurately. Across experimental conditions, Fast-dm was used to estimate drift rate (v), boundary separation (a), and the nondecision component (Ter) for each participant. For

5

Figure 2. Lexical decision performance from Experiment 1 as a function of Relatedness and Quantiles in the legal nonword condition (top panel) and transposed-letter (TL) nonword condition (bottom panel). Empirical quantiles are represented by error bars, whereas fitted ex-Gaussian quantiles are represented by lines. RT ⫽ response time.

v, larger numbers reflect steeper drift rates; for a, larger numbers reflect more conservative response criteria; and for Ter, larger numbers reflect longer nondecision times. For v, only the main effect of nonword type was significant, F(1, 78) ⫽ 30.81, p ⬍ .001, MSE ⫽ 0.751, ␩2p ⫽ .28; none of the other effects were reliable, Fs ⬍ 1. For a, only the main effect of nonword type was significant, F(1, 78) ⫽ 29.69, p ⬍ .001, MSE ⫽ 0.200, ␩2p ⫽ .28; none of the other effects were reliable, Fs ⬍ 1. For Ter, only the main effect of priming was significant, F(1, 78) ⫽ 11.43, p ⫽ .001, MSE ⫽ 0.003, ␩2p ⫽ .13; none of the other effects were reliable, ps ⬎ .25. Summary. Although the presence of TL nonword distracters substantially slowed responses to word targets, priming effects were not moderated by nonword type, replicating Lupker and Pexman (2010). Furthermore, distributional analyses revealed that priming—for both nonword types—was reflected only by distributional shifting (␮), whereas nonword type effects—for related and unrelated targets—were reflected by shifting (␮) and an increase in the tail (␶) of the distribution. Finally, the diffusion model analyses indicated that the distributional shift associated with priming was mediated entirely by nondecision time (Ter; i.e., the time taken for encoding and response execution), not by drift rate (v) or boundary separation (a). In contrast, the effect of nonword type was reflected in lower drift rates and more conservative response criteria when TL nonwords were used as distracters.


6

Results and Discussion

Figure 3. Priming effect across quantiles as a function of nonword type. Error bars reflect the standard errors of the difference scores. TL ⫽ transposed-letter.

Experiment 2 Method Participants. Forty-eight participants (30 females) from the National University of Singapore participated for course credit. All participants were proficient in English and had normal or corrected-to-normal vision. The mean vocabulary age, as measured by Shipley’s (1940) Vocabulary subscale, was 17.90 (SD ⫽ 1.11). Design. Priming (related or unrelated) and Stimulus Quality (clear or degraded) were manipulated within-participants. The dependent variables were RT and accuracy. Stimuli. The 120 prime-target pairs and legal nonwords from Experiment 1 were used. Procedure. The procedure of Experiment 2 was similar to that used in Experiment 1, except that for half the trials (i.e., the degraded condition), letter strings were rapidly alternated with a randomly generated mask of the same length. For example, the mask @$#&% was presented for 14 ms, followed by a five-letter target word for 28 ms, and the two repeatedly alternated until the participant responded. The masks were generated from random permutations of the following symbols: &@?!$*%#?⬎. Across participants, targets were counterbalanced across related and unrelated conditions and degraded and clear conditions.

Errors (6.3% across all conditions) and response latencies shorter than 200 ms or longer than 3,000 ms were excluded from the analyses. Response latencies beyond 2.5 SDs from each participant’s mean were also excluded, removing an additional 3.4% of the responses. The mean RTs, accuracy rates, and ex-Gaussian parameters are presented in Table 3. Response latencies. For RTs, the main effects of stimulus quality, Fp(1, 47) ⫽ 96.55, p ⬍ .001, MSE ⫽ 9,791, ␩2p ⫽ .67; Fi(1, 119) ⫽ 486.21, p ⬍ .001, MSE ⫽ 5,123, ␩2p ⫽ .80, and priming, Fp(1, 47) ⫽ 35.37, p ⬍ .001, MSE ⫽ 2,582, ␩2p ⫽ .43; Fi(1, 119) ⫽ 25.23, p ⬍ .001, MSE ⫽ 9,222, ␩2p ⫽ .18, were significant. The Stimulus Quality ⫻ Priming interaction was also significant, Fp(1, 47) ⫽ 8.92, p ⫽ .004, MSE ⫽ 2,467, ␩2p ⫽ .16; Fi(1, 119) ⫽ 12.05, p ⫽ .001, MSE ⫽ 6,420, ␩2p ⫽ .09. Priming effects were larger for degraded (65 ms) than for clear (22 ms) targets. Accuracy rates. For accuracy rates, the main effects of stimulus quality, Fp(1, 47) ⫽ 29.70, p ⬍ .001, MSE ⫽ 0.002, ␩2p ⫽ .39; Fi(1, 119) ⫽ 11.55, p ⫽ .001, MSE ⫽ 0.015, ␩2p ⫽ .088, and priming, Fp(1, 47) ⫽ 12.81, p ⫽ .001, MSE ⫽ 0.002, ␩2p ⫽ .21; Fi(1, 119) ⫽ 11.76, p ⫽ .001, MSE ⫽ 0.006, ␩2p ⫽ .090, were significant. The Stimulus Quality ⫻ Priming interaction approached significance by participants and by items, Fp(1, 47) ⫽ 2.99, p ⫽ .090, MSE ⫽ 0.002, ␩2p ⫽ .06; Fi(1, 119) ⫽ 3.38, p ⫽ .069, MSE ⫽ 0.005, ␩2p ⫽ .028. Priming effects were larger for degraded (3.6%) than for clear (1.3%) targets. Ex-Gaussian analyses. For ␮, the main effects of stimulus quality, Fp(1, 47) ⫽ 83.81, p ⬍ .001, MSE ⫽ 3,021, ␩2p ⫽ .64, and priming, Fp(1, 47) ⫽ 13.05, p ⫽ .001, MSE ⫽ 2,599, ␩2p ⫽ .22, were significant. The Stimulus Quality ⫻ Priming interaction was not significant, F ⬍ 1. Turning to ␴, none of the effects were reliable, Fs ⬍ 1. Finally, for ␶, the main effect of stimulus quality was significant, F(1, 47) ⫽ 20.16, p ⬍ .001, MSE ⫽ 11,001, ␩2p ⫽ .30, but the main effect of priming did not reach significance (p ⫽ .096). This was qualified by a borderline reliable Stimulus Quality ⫻ Priming interaction, F(1, 47) ⫽ 3.60, p ⫽ .064, MSE ⫽ 3,162, ␩2p ⫽ .07, where priming effects were larger for degraded (30 ms) than for clear (⫺1 ms) targets. Quantile analyses. The mean quantiles for the different experimental conditions are plotted in Figure 4, whereas Figure 5 presents priming effects as a function of stimulus quality. From

Table 3 Mean RTs, Accuracy, Ex-Gaussian, and Diffusion Model Parameters as a Function of Stimulus Quality and Relatedness in Experiment 2 Stimulus quality/priming Clear Related Unrelated Priming effect Degraded Related Unrelated Priming effect Interaction Note. RT ⫽ response time.

M

Accuracy

␮

␴

␶

v

a

Ter

650 672 22

0.97 0.96 0.01

508 531 23

55 61 6

145 144 ⫺1

2.78 2.88 ⫺0.10

1.34 1.34 0

0.43 0.45 0.02

769 834 65 43

0.94 0.91 0.03 0.02

577 608 31 8

63 61 ⫺2 ⫺8

198 228 30 31

2.50 2.23 0.27 0.37

1.47 1.54 0.07 0.07

0.51 0.53 0.02 0


7

reflected by nondecision time (Ter), consistent with Experiment 1. However, when targets were degraded, priming was mediated by nondecision time and drift rate (v); compared to related targets, unrelated targets were associated with reliably lower drift rates.

General Discussion

Figure 4. Lexical decision performance from Experiment 2 as a function of Relatedness and Quantiles in the clear condition (top panel) and degraded condition (bottom panel). Empirical quantiles are represented by error bars, whereas fitted ex-Gaussian quantiles are represented by lines. RT ⫽ response time.

Figure 4, it is clear that priming effects for clear targets remained relatively constant across the quantiles, but priming effects for degraded targets increased across the RT distribution. Diffusion model analyses. For v, the main effect of stimulus quality, F(1, 47) ⫽ 53.05, p ⬍ .001, MSE ⫽ 0.200, ␩2p ⫽ .53, was significant. The Stimulus Quality ⫻ Priming interaction was also significant, F(1, 47) ⫽ 8.02, p ⫽ .007, MSE ⫽ 0.200, ␩2p ⫽ .15; priming effects were reliable for degraded (p ⫽ .002) but not for clear (p ⫽ .379) targets. Moreover, degradation effects were reliable for both related and unrelated targets (ps ⬍ .01), with stronger degradation effects observed for unprimed words. Turning to a, only the main effect of stimulus quality was significant, F(1, 47) ⫽ 19.09, p ⬍ .001, MSE ⫽ 0.069, ␩2p ⫽ .29; none of the other effects were reliable, ps ⬎ .19. For Ter, the main effects of stimulus quality, F(1, 47) ⫽ 109.07, p ⬍ .001, MSE ⫽ 0.002, ␩2p ⫽ .70, and priming, F(1, 47) ⫽ 5.95, p ⫽ .019, MSE ⫽ 0.004, ␩2p ⫽ .11, were significant; the Stimulus Quality ⫻ Priming interaction was not reliable, F ⬍ 1. Summary. Priming effects were larger for degraded, compared to clear, words, and this interaction was predominantly mediated by ␶, the tail of the distribution, consistent with the extant literature (see, e.g., Balota et al., 2008). Specifically, priming reflected a shift (␮) for clear targets, but shifting (␮) and an increase in the tail (␶) for degraded targets. The diffusion model analyses revealed that for clear targets, priming was primarily

In the present study, we explored how the well-studied semantic priming effect was moderated by two variables known to robustly increase lexical decision difficulty: nonword type (TL nonwords vs. legal nonwords) and stimulus quality (clear vs. degraded). In order to increase the clarity of our comparisons, we used a common set of prime-target pairs, and participants across different experiments were sampled from the same participant pool and were comparable on vocabulary knowledge. It is first worth noting that both manipulations strongly increased discrimination difficulty. Specifically, in Experiment 1, TL nonword distracters (e.g., JUGDE), compared to legal nonwords, slowed RTs to words by 158 ms. In Experiment 2, visually degrading words slowed RTs by 115 ms. Importantly, there was a dissociation between the effects of nonword type and stimulus quality on semantic priming. While nonword type and priming produced additive effects (replicating Lupker & Pexman, 2010), stimulus quality and priming interacted, yielding greater priming for degraded words. More intriguingly, distributional analyses revealed that semantic priming was reflected by shifting of comparable magnitude for legal and TL nonwords (see Figure 3). In contrast, although priming for clear targets was reflected by a shift, priming for degraded targets reflected both shifting and an increase in the tail of the distribution (see Figure 5). Supplementary diffusion model analyses also indicated that the distributional shift associated with the priming of clear targets is most consistent with changes in nondecision time, whereas the priming of degraded targets implicates changes in nondecision time and, importantly, drift rate. Although the main effects of nonword type and stimulus quality were of secondary interest in the present study, the distributional analyses also yielded insights into how these factors influence word recognition performance. Specifically, the slowing observed in the context of difficult TL nonwords was mediated more strongly by ␶ (i.e., the tail of the distribution) than by ␮, consistent with Yap et al. (2006). The presence of TL nonwords was also associated with lower drift rates and more conservative response

Figure 5. Priming effect across quantiles as a function of stimulus quality. Error bars reflect the standard errors of the difference scores.

8


criteria but had no influence on nondecision time. Turning to stimulus quality effects, the slowing afforded by degrading the stimulus was reflected to a similar extent in ␮ and ␶ (replicating Yap & Balota, 2007). In addition, degrading words resulted in lower drift rates, longer nondecision times, and more conservative response criteria. We are not aware of any theoretical framework or model that has attempted to provide a unified explanation for the dissociations reported in the present study, along with other well-established findings in the lexical decision literature. In the remainder of this article, we discuss in greater depth how these results might be accommodated.

Joint Effects of Stimulus Quality, Priming, and Nonword Type: Going Beyond the Mean In the present study, we examined the targeted effects both at the level of the mean and at the level of underlying RT distributional characteristics. In order to better characterize distributional effects, data were examined using a combination of ex-Gaussian, quantile, and diffusion model techniques. The results from Experiment 1 indicated that semantic priming was reflected by distributional shifting of comparable magnitude for clearly presented targets, whether legal or TL nonwords were used (see Figure 3). In terms of diffusion model parameters, this semantic priming shift was mediated by nondecision time but not by drift rate or response boundary. However, turning to Experiment 2, a different pattern was found. Specifically, when targets were degraded, priming was reflected by both shifting and an increase in the tail of the distribution (see Figure 5), consistent with an increase in nondecision time and drift rate. The results from the diffusion model, which captures the distributional changes, are informative in a number of ways. First, the distributional shift associated with priming clear targets largely reflects an effect on the duration of the nondecision component, which taps processes involved in stimulus encoding and response execution. Second, drift rate mediates the priming effect only when targets are degraded. Both of these patterns would appear to place further constraints on Ratcliff and McKoon’s (1988) compoundcue model of priming, which emphasizes changes in drift rate across related and unrelated primes. Finally, whereas increasing difficulty via nonword type reliably decreased word drift rate and made participants more conservative, priming was equivalent across different types of nonwords, both at the level of the mean and at the level of distributional characteristics. Balota et al. (2008) have argued that priming that is reflected by a shift (observed in highly skilled readers) is most consistent with a relatively modular headstart mechanism, wherein primes produce a constant amount of preactivation for the logogens (Morton, 1969) of related targets. This notion of a headstart meshes well with the finding that priming influences the nondecision (i.e., encoding and response execution) component rather than the response boundaries of a diffusion process (cf. Borowsky & Besner, 1993). It is worth noting that adjusting response boundaries can also produce distributional shifting in a simple random walk model (Spieler, Balota, & Faust, 2000; Yap et al., 2006), so it is intriguing that the diffusion model analyses attribute shifting to nondecision time. Distributional shifting cannot be easily reconciled with feature-overlap models of priming (Masson, 1995; Plaut & Booth,

2000), which predict changes in the scale of the RT distribution (i.e., slower trials should be associated with more priming). As noted, the compound-cue model also assumes that priming effects are generally mediated by changes in drift rate, which is of course inconsistent with the present observation that priming as reflected by drift rate variations is observed only for degraded targets. When targets are visually degraded, priming is reflected by both shifting and an increase in the tail of the distribution. Consistent with this, supplementary diffusion model analyses revealed that priming for degraded targets involved changes in nondecision time and drift rate. To explain this, Balota et al. (2008) have proposed that degraded targets are difficult to resolve, and the lexical system therefore adaptively retrieves more prime information when processing such targets (for an adaptive rational analysis of semantic priming, also see Anderson & Milson, 1989; Bodner, Masson, & Richard, 2006). Moreover, the distributional plots (see Figure 5) indicate that reliance on prime information is proportional to the difficulty of the trial. Therefore, the more difficult items—that is, those at the slow end of the RT distribution—are associated with more priming. This suggests that when a target is degraded, priming reflects both a headstart and an additional prime retrieval mechanism that is engaged after the target is presented and which is flexibly modulated by the fluency of target processing (see Balota & Yap, 2006, for a discussion of flexible lexical processing). The distinction between priming as a headstart and priming as a flexible prime retrieval process aligns nicely with the distinction between prospective versus retrospective priming mechanisms (Thomas et al., 2012). Prospective priming refers to the facilitation afforded by a prime before a target is presented (e.g., Balota et al., 2008), whereas retrospective priming (e.g., Balota et al., 2008; Bodner & Masson, 1997; Neely, Keefe, & Ross, 1989; Ratcliff & McKoon, 1988) implicates processes that operate after the target is presented. In this light, distributional shifting might be viewed as a marker for prospective priming mechanisms, whereas an increase in the distribution’s tail may reflect a more strategic retrospective mechanism. Interestingly, Thomas et al. (2012) recently explored the effect of prime-target associative direction on the Stimulus Quality ⫻ Priming interaction. Prospective priming mechanisms necessarily depend on the presence of a forward prime-to-target association (e.g., KEG– BEER) and cannot account for backward priming (i.e., target-toprime association; e.g., SMALL–SHRINK). Thomas et al. reported that the overadditive interaction between stimulus quality and priming was seen for trials with symmetric (e.g., EAST–WEST) or backward (target-to-prime) associations but was not seen for trials with only forward (prime-to-target) associations. This suggests that the Stimulus Quality ⫻ Priming interaction is entirely mediated by a retrospective prime retrieval mechanism that depends on the presence of a backward target-to-prime association. We comment more on this intriguing finding later.

Dissociating Nonword Type and Stimulus Quality Effects: Implications for Models We now consider the broader theoretical implications of our findings for models of word recognition and lexical decision performance. The present study was motivated by an intriguing dissociation reported by Lupker and Pexman (2010), wherein the joint effects of nonword type and frequency were very different from the joint effects of nonword type and semantic priming. That


is, whereas frequency effects became larger in the presence of TL nonwords, priming effects remained the same size. According to Lupker and Pexman, any model that accounts for nonword type effects in lexical decision performance using a single, lexically driven mechanism is hard pressed to account for this pattern. These include the MROM (Grainger & Jacobs, 1996), the diffusion model (Ratcliff et al., 2004), and the Bayesian reader model (Norris, 2006, 2009). First, let us consider how the three models capture the frequency by nonword type interactions. In the MROM, lexical decisions to words are made when the activation level of individual lexical representations (i.e., local activity) or the summed activation of all representations (i.e., global activity) exceed their respective thresholds. The response threshold for global activation, also known as the ␴ criterion, is set higher as nonword distracters become more wordlike. This implies that as nonword wordlikeness increases, word responses are influenced to a greater extent by local, relative to global, lexical activity, resulting in longer latencies. Importantly, variations in the ␴ criterion (as a function of nonword type) exert more influence on low-frequency than on high-frequency words, yielding larger frequency effects in the context of wordlike nonwords. Turning to the diffusion model (Ratcliff et al., 2004), nonword context has a greater impact on the drift rate of lowfrequency words than on the drift rate of high-frequency words. Consequently, the difference between the drift rates of high- and low-frequency words is magnified as distracters become more wordlike. Finally, in the Bayesian reader model (Norris, 2006, 2009), the model continuously computes the ratio of the summed likelihood that the presented letter string is a word to the summed likelihood that it is a nonword; evidence for a word response accumulates more rapidly as this ratio becomes larger. Increasing distracter wordlikeness increases the nonword likelihood, which slows down lexical decision to both words and nonwords. Indeed, the Bayesian reader successfully simulates the Nonword Type ⫻ Word-Frequency interaction (see Norris, 2009, for more details). In short, the three single-process models described above can accommodate the well-known interaction between nonword type and word-frequency. With the exception of the diffusion model, which accounts for word-frequency (Ratcliff et al., 2004) and priming (Ratcliff & McKoon, 1988) effects via drift rate changes, the other models have not yet been extended to explicitly account for semantic priming effects. However, priming effects could, in principle, reflect the same mechanisms underlying frequency effects. Specifically, just as activation thresholds (or resting activation levels) are lower for high-frequency words, compared to low-frequency words, activation thresholds (or resting activation levels) could be lower for targets primed by related words, compared to targets primed by unrelated words. However, as Lupker and Pexman (2010) have correctly pointed out, if priming and frequency effects indeed implicate a common lexical mechanism, then semantic priming should also interact with nonword type, a prediction that is clearly at odds with their findings and the results of Experiment 1. In addition to the problems posed for the extant models by the differential effect of nonword difficulty on semantic priming and word-frequency effects, one finds a qualitatively different pattern when one considers increasing lexical decision difficulty by degrading the target. The Stimulus Quality ⫻ Priming interaction (explored in Experiment 2), which at the level of the mean has been replicated

9

by various investigators (e.g., Balota et al., 2008; Borowsky & Besner, 1993; Brown & Besner, 2002; Meyer et al., 1975; Thomas et al., 2012), in conjunction with the well-known additive effects of stimulus quality and frequency (Balota & Abrams, 1995; Becker & Killion, 1977; Plourde & Besner, 1997; Stanners, Jastrzembski, & Westbrook, 1975; Yap & Balota, 2007) provide an interesting counterpoint to the intriguing relationships between nonword type, frequency, and priming. Remarkably, nonword type is additive with priming but interactive with frequency, whereas stimulus quality is additive with frequency but interactive with priming. Again, if one assumes that frequency and priming effects implicate a common lexical locus, then it is not obvious why stimulus quality is additive with the former and interactive with the latter. Possibly, one needs to consider a completely different perspective to account for this complex pattern of results. Plaut and Booth (2000) proposed a single-mechanism parallel distributed processing (PDP) model, in which input and output processes are mediated by a nonlinear sigmoid function. The full details of this model are beyond the scope of this report, but essentially, the nonlinear function allows equal differences in the input to be reflected by equal or unequal differences of the output, depending on the portion of the function being considered (see Figure 6). Words and nonwords are discriminated on the basis of semantic stress (a measure of familiarity); words, compared to nonwords, are associated with higher stress values. Hence, this connectionist model, unlike the single-process models described above, can potentially accommodate the complex joint effects of stimulus quality, wordfrequency, and semantic priming. Although Plaut and Booth have explicitly mentioned that the specific implemented model they

Figure 6. The sigmoid activation function of Plaut and Booth’s (2000) model. HF ⫽ high-frequency; H ⫽ high; LF ⫽ low-frequency; L ⫽ low; Unrel ⫽ unrelated; Rel ⫽ related; Freq ⫽ frequency. Reproduced with permission from “Individual and Developmental Differences in Semantic Priming: Empirical and Computational Support for a Single-Mechanism Account of Lexical Processing,” by D. C. Plaut and J. R. Booth, 2000, Psychological Review, 107, p. 832. Copyright 2000 by the American Psychological Association.


10

described could not handle nonword type effects (see p. 812 of their article), an earlier PDP model (Plaut, 1997) successfully simulated nonword type effects in isolated word recognition by implementing the assumption that more wordlike nonwords (such as pseudohomophones or TL nonwords) should be associated with higher stress values than less wordlike nonwords. However, it is noteworthy that the architectures of the two models are quite different. Specifically, Plaut’s (1997) model, compared to its successor, had twice as many semantic units, had 10 times the number of hidden units, had six times the number of orthographic units, and was feedforward rather than recurrent (see Borowsky & Besner, 2006, for more details). Whether Plaut and Booth’s model can indeed discriminate between words and wordlike nonwords accurately while at the same time correctly simulating the joint effects of nonword type, stimulus quality, frequency, and priming is an interesting empirical question that can only be appropriately addressed via simulations on the actual implemented model. Certainly, it will be challenging for the model to explain why manipulating the semantic stress of the nonword distracters exerts such different influences on semantic priming and word-frequency effects. Additionally, Borowsky and Besner (2006; Besner & Borowsky, 2006) have highlighted how the PDP model’s reliance on a sigmoid function to relate input activation to RTs does not allow the model to simulate additive effects of two factors (i.e., nonword type and priming) when one of those factors (i.e., nonword type) produces an interaction with a third factor (i.e., wordfrequency) within the same range of RTs (see also Plaut & Booth, 2006). A more recent study by Besner, Wartak, and Robidoux (2008) has also indicated that the model incorrectly simulates the joint effects of stimulus quality and word-frequency. Specifically, the model produces underadditivity, additivity, or overadditivity of the two factors depending on the size of the stimulus quality effect, whereas skilled readers yield a broader pattern of additivity.

Tying It All Together: The Case for a Multistage Model The foregoing discussion makes it clear that even if one only considers mean-level RT performance, single-process models that postulate a common locus for frequency and priming effects will have difficulty explaining why (1) nonword type is additive with priming but interactive with frequency and (2) stimulus quality is

additive with frequency but interactive with priming. McNamara (2005) has also pointed out that traditional priming accounts (e.g., spreading activation, verification, compound-cue) are too simple to capture the complex relations among word-frequency, stimulus quality, and semantic priming, and we have now added the differential effects of stimulus quality and nonword type on semantic priming to the mix (see Table 4). As a starting point in this discussion, we are relying on Sternberg’s (1969) additive factors logic to interpret the pattern of RT factor effects observed in the present study and to make inferences about the organization of the lexical processing architecture. The logic is based on the premise that independently changeable, serially arranged stages underlie lexical processing, and that processing is discrete (i.e., thresholded), that is, processing in a later stage begins only after processing in an earlier stage is complete. Additive effects of factors suggest that they influence different stages, whereas interactive effects suggest that they influence at least one common stage. Because the current discussion relies on additive factors logic, it is important to remember there are some limitations. Specifically, it has been known for some time that although independent, separately modifiable stages imply additive effects, additive effects do not necessarily imply separate stages. For example, in the cascade model (Ashby, 1982; McClelland, 1979), all processes are operating continuously, and information is passed from one process to the next as soon it becomes available. Importantly, the cascade model, under some parameter constraints, is also able to produce approximately additive effects without a discrete-stage architecture. In order to better adjudicate between the two alternatives, Roberts and Sternberg (1993) examined the extent to which the stage and cascade models could account for human performance at the level of the mean RT and, more importantly, at the level of the RT distributions across different experiments (detection, identification, classification). Overall, they found that the predictions from the stage model provided a better fit to the empirical data sets. Interestingly, the cascade model was most successful under parameter settings where it resembled a stage model, although even this version of the cascade model was rejected after it made incorrect predictions about the relations among means and variances (Roberts & Sternberg, 1993; Sternberg, 1998). Related to this, Yap and Balota (2007) also observed

Table 4 Joint Effects of Stimulus Quality, Priming, Word-Frequency, and Nonword Type Effect 1. Stimulus Quality ⫻ Priming ⫻ RP 2. Priming ⫻ Word-Frequency ⫻ Vocabulary

3. Nonword Type ⫻ Word-Frequency

Description of interaction (if applicable) Additive effects of Stimulus Quality and Priming when RP is low (.25). More priming for degraded, compared to clear, targets when RP is high (.50). Additive effects of Priming and Word-Frequency for high-vocabulary participants. More priming for LF, compared to HF, targets for the low-vocabulary participants. Larger word-frequency effects when more wordlike nonwords are used.

4. Additive effects of Stimulus Quality and Word-Frequency 5. Additive effects of Nonword Type and Priming Note. RP ⫽ relatedness proportion; LF ⫽ low-frequency; HF ⫽ high-frequency.

Study Stolz & Neely (1995) Yap et al. (2009)

Stone & Van Orden (1993) Yap & Balota (2007) Lupker & Pexman (2010)


11

additive effects of stimulus quality and word-frequency in means and in higher order moments (i.e., variance and skewness). This is a challenging pattern for the cascade model to handle, because it predicts additivity at the level of the mean but not necessarily at the level of higher order moments (Roberts & Sternberg, 1993). At a more general level, it is unclear how any model based on the interactive activation framework (which is predicated on top-down and bottom-up cascaded processing between modules) can handle the combined pattern of additive and interactive effects in the present study. For these reasons, and given the current level of development of computational models in the field, we believe that the stage framework continues to be a useful, albeit metaphorical, way for accommodating additive and interactive patterns of results from factorial designs.

Multistage Processing in Lexical Decision In the remainder of this article, we describe how a multistage framework can provide an understanding of the intriguing qualitative differences between stimulus quality and nonword type effects, and between priming and word-frequency effects. To preview, our descriptive model is predicated on the idea that multiple stages of processing subserve lexical-semantic processing (Borowsky & Besner, 1993, 2006; Stolz & Besner, 1998). This allows the loci of priming and word-frequency effects to be further decoupled, along with an attempt to understand the clear differences in the underlying RT distributions as a function of these variables. Like Borowsky and Besner (1993), we do not make precise claims about the specific nature of the activation (i.e., discrete vs. continuous) across stages, but for ease of exposition, it is assumed that the model can contain both discrete and continuous processing features (see McNamara, 2005). Importantly, it has been shown that additive factors logic can be extended to cascaded processing frameworks or hybrid frameworks that include discrete and continuous features (McClelland, 1979, p. 312). Of course, we acknowledge that aspects of this model are speculative. However, as one reviewer has pointed out, it may take a while before word recognition researchers are able to develop a cogent, integrated account of the full range of extant data. In the meantime, we believe that a staged framework provides a useful way for organizing our findings and the attendant theorizing. As a starting point, we adopt the multistage framework suggested by McNamara (2005) and Stolz and Besner (1998). According to this framework, there are three levels (letter, lexical, semantic) of representation and processing (see Figure 7); connections between different levels are excitatory, whereas connections within levels are inhibitory. Presenting a prime (e.g., NURSE) first activates letter-level representations, followed by activation at the lexical (via Pathway A) and semantic (via Pathway B) levels. Semantic priming is the result of spreading activation from the target word (i.e., NURSE) to related words (e.g., DOCTOR) within the semantic level. At the same time, activation is feeding backward from the semantic level to the lexical (via Pathway C) and letter (via Pathway D) levels, respectively preactivating the lexical-level and letter-level representations for DOCTOR. Hence, should DOCTOR appear shortly after NURSE, it can be processed more efficiently because its letter-, lexical-, and semantic-level representations have been partially activated. For our purposes, two additional points are worth highlighting. Based on the additive

Figure 7. An interactive-activation multistage model of semantic priming. Reprinted with permission from Semantic Priming: Perspectives From Memory and Word Recognition, by T. P. McNamara, 2005, p. 41, New York, NY: Psychology Press. Copyright 2005 by the Taylor & Francis Group.

effects of stimulus quality and frequency (Yap & Balota, 2007) and the additive effects of priming and frequency for skilled lexical processors (Yap et al., 2009), it is assumed that stimulus quality influences only the letter level directly, word-frequency influences only the lexical level directly, and semantic priming influences only the semantic level directly. The framework also does not possess a dedicated mechanism for making lexical decisions. As we discuss in a later section, this omission has important implications. How might this model account for extant effects, both at the level of mean RTs and at the level of RT distributional characteristics? First, consider the interaction between stimulus quality and priming, where priming effects are larger for degraded targets. The traditional account is as follows: Semantic priming reflects spreading activation within the semantic level and feedback to lower levels, which collectively preactivates the semantic, lexical, and letter-level representations of targets related to the prime. As a consequence of this feedback, related targets will be disrupted less by visual degradation than unrelated targets, yielding an overadditive interaction. Note that both spreading activation and semantic feedback are prospective in nature, because these processes occur before the target is presented. In other words, within the multistage framework, the received explanation for the Stimulus Quality ⫻ Priming interaction is based solely on prospective priming mechanisms. As discussed earlier, the recent study by Thomas et al. (2012) places further constraints on the mechanisms driving this interaction. Specifically, their findings clearly indicate that the Stimulus Quality ⫻ Priming interaction is mediated by a retrospective prime-retrieval mechanism that relies on backward targetto-prime associations. This mechanism is strategically engaged

12


when target processing is difficult (e.g., when degraded or lowfrequency words are presented), so that prime information can compensate for compromised bottom-up processing. Hence, instead of semantic feedback, the greater influence of priming for degraded targets might be better attributed to a retrospective prime retrieval process (e.g., retrospective semantic-matching; Neely & Keefe, 1989). In this light, we can now revisit the interaction between stimulus quality and priming, where priming is reflected by distributional shifting for clear targets but increases across quantiles for degraded targets. When a prime is presented (e.g., NURSE), related representations in the semantic level1 (e.g., DOCTOR) are preactivated through spreading activation, affording them a headstart on target processing. In this instance, the headstart yielded by the related prime is prospective in nature, because the facilitation occurs before the target is presented (e.g., Balota et al., 2008). However, when targets are difficult to resolve in some way (e.g., they are visually degraded), this engages a prime-retrieval mechanism (Thomas et al., 2012). The operation of this mechanism is reflected by both shifting and an increase in the slow tail of the distribution. Importantly, the influence of the prime is proportional to the difficulty of the trial, that is, priming effects are larger for slower (presumably more difficult) targets when they are degraded. Prime retrieval is a retrospective mechanism (e.g., Balota et al., 2008; Bodner & Masson, 1997; Neely et al., 1989; Ratcliff & McKoon, 1988) because it implicates processes after the target is presented. The dichotomy between prospective (headstart) and retrospective (prime retrieval) priming is consistent with other empirical work. For example, Stolz and Neely (1995) demonstrated that the Stimulus Quality ⫻ Priming interaction was present only when relatedness proportion (RP; i.e., the proportion of word targets preceded by a related prime) was high (e.g., 0.50). When relatedness proportion was decreased (e.g., 0.25), the two variables produced additive effects. In other words, retrospective prime retrieval, which the Stimulus Quality ⫻ Priming interaction is a marker for, occurs only in the high RP condition (i.e., when the utility of a related prime is high). Related to this, Yap et al. (2009) reported that the Frequency ⫻ Priming interaction was observed only in participants with relatively lower levels of vocabulary knowledge. Participants with more vocabulary knowledge produced additive effects of the two factors. This fits well with the aforementioned proposal that there is retrospective prime retrieval only when the utility of related prime information is high. Specifically, prime retrieval is not evident for higher-vocabularyknowledge participants because the utility of the prime is relatively low for such participants who, by virtue of possessing highintegrity lexical representations (e.g., Perfetti & Hart, 2002), are able to process both high- and low-frequency targets fluently. In summary, the RT distributional results of the Priming ⫻ Stimulus Quality interaction are consistent with the notion that there is retrospective prime retrieval when target processing is effortful and the utility of the prime is high (see Thomas et al., 2012, for further discussion of prospective and retrospective priming effects). We believe the multistage framework proposed above can also be extended in a relatively straightforward manner to accommodate the additive effects of nonword type and priming. As discussed earlier, the Nonword Type ⫻ Word-Frequency interaction,

in conjunction with the additive effects of nonword type and priming, cannot be reconciled with the premise that nonword type, word-frequency, and semantic priming affect the same, lexicallydriven mechanism (Lupker & Pexman, 2010). Instead, one needs to posit additional processes/mechanisms, which are more tuned to the task-specific decision mechanisms in the lexical decision task. Specifically, while a primary set of lexical processes is sensitive to frequency and priming (but not to nonword type), a secondary process is influenced by frequency and nonword type (but not by priming). In this light, we suggest augmenting the multistage framework with an additional word/nonword discrimination mechanism that operates after the semantic level. Indeed, incorporating such a decision-making stage is compatible with the idea that in lexical decision, experimental variables (e.g., word-frequency) can influence both lexical processing and postlexical decision-making mechanisms that are specific to the task (Balota & Chumbley, 1984). For example, the decision stage could be modeled after familiarity-based accounts (e.g., Balota & Chumbley, 1984; Besner, 1983; Besner & Swan, 1982), which have been useful for explaining nonword type effects (e.g., Yap et al., 2006) and other lexical decision phenomena. Familiarity-based accounts strongly emphasize the role of any information that can be recruited by participants to discriminate between words and nonwords. Because words are more familiar and meaningful than nonwords, familiarity2 (and any variable correlated with familiarity) is a useful dimension for driving the discrimination process. For example, frequency influences the postlexical discrimination process because low-frequency words, compared to high-frequency words, are more similar to nonwords on the relevant familiarity/meaningfulness dimension. Because of the increased overlap between low-frequency words and nonwords, participants may be compelled to engage in more attentiondemanding analyses (e.g., spell-checking or retrieving semantic referent) for such words, hence slowing lexical decision latencies (Balota & Spieler, 1999). Clearly, in the context of very wordlike nonwords that include a transposed pair of letters, it would be adaptive for the decision maker to conduct a check process to ensure that the spelling is correct. Alternatively, the postlexical decision stage might be represented by a diffusion-based process (Ratcliff et al., 2004), which is able to account for correct and error RTs very well, both at the level of the mean and at the level of their distributions. Specifically, during target processing, the system is monitoring any activity that is relevant for discriminating between words and nonwords, and this activity could map on to a wordness value (i.e., a familiarity-like metric that reflects how wordlike a stimulus is) that drives drift rates in a noisy diffusion decision process (Plaut, 1997; Ratcliff et al., 2004). In order to accommodate nonword type effects, the mapping between semantic activity and drift rate could be differentially weighted depending on the extent of overlap between words and nonword distracters (see Yap et al., 2006, for 1 Although we are suggesting that semantic priming truly resides in the semantic memory system, there is evidence that the priming effect may indeed be due to associative co-occurrence and, hence, resides within the lexical network (see review by Hutchison, 2003). 2 Here, familiarity refers to a multidimensional quantity that reflects the orthographic and phonological similarity of a letter string to real words.


more discussion). Of course, without explicit modeling of the full set of effects, there is no assurance that a hybrid model that integrates multiple stages with a diffusion mechanism can faithfully approximate performance. To recapitulate, the additive effects of nonword type and priming are consistent with the notion that nonword type, but not priming, influences postlexical decision-making mechanisms that are sensitive to familiarity or wordness. Of course, this predicts that variables that contribute to the familiarity/wordness of a letter string should interact with nonword type. The canonical example is word-frequency, that is, high-frequency words are perceived as more familiar than low-frequency words. Other variables that could potentially increase familiarity/wordness include imageability (Cortese & Fugett, 2004), age of acquisition (Cortese & Khanna, 2007; Juhasz, 2005), number of senses (Pexman & Lupker, 1999), number of associates (i.e., number of distinct first associates elicited in a free association task; Dun˜abeitia, Avile´s, & Carreiras, 2008), number of features (Pexman, Lupker, & Hino, 2002), body-object interaction (i.e., the extent to which referents can be physically interacted with; Siakaluk, Pexman, Aguilera, Owen, & Sears, 2008), sensory experience rating (i.e., the extent to which a word evokes sensory/perceptual experiences; Juhasz, Yap, Dicke, Taylor, & Gullick, 2011), and repetition (Balota & Spieler, 1999). With this in mind, it is noteworthy that nonword type has already been shown to interact with frequency (Stone & Van Orden, 1993), imageability (James, 1975), number of senses (Pexman & Lupker, 1999), and number of features (Pexman et al., 2002), with the other variables yet to be investigated. Of course, the corollary here is that priming a word with its semantic associate does not increase its familiarity/wordness. Why does semantic priming not increase the familiarity or wordness of a word? The repetition priming (i.e., primes are identical to the target) literature might be informative here. Although the precise bases of repetition priming remain controversial (see Tenpenny, 1995, for a review), most researchers agree that repetition effects involve multiple mechanisms, that is, a shortterm effect reflecting lexical activation and a long-term effect reflecting memory trace retrieval (see Coane & Balota, 2011; Versace & Nevers, 2003). Interestingly, Forster and Davis (1984) have suggested that the locus of repetition priming’s episodic influence, but not its short-term lexical effect, is at the decision stage of the lexical decision task. We have proposed that in the absence of retrospective prime retrieval, which appears to be modulated by the utility of the prime, semantic priming effects reflect a simple headstart mechanism, wherein primes preactivate related targets to the same extent. Parenthetically, the notion of a headstart converges nicely with the observation that the semantic priming of clear targets reflects changes in nondecision (i.e., encoding and response execution) time but not drift rate. Assuming that these preactivated targets do not lay down episodic traces, and priming-based familiarity/wordness critically depends on episodic influences, it therefore follows that headstartmediated semantic priming should produce additive effects with any variable (e.g., nonword type) whose influence is localized to the decision stage. Indeed, the notion that semantic and repetition priming tap different processes meshes well with den Heyer and Goring’s (1985) finding that the effects of semantic and repetition priming are additive in lexical decision. Of course, it is important

13

to emphasize that this explanation is post hoc and needs to be empirically verified in future studies.

Conclusions In summary, semantic priming is one of the most robust and extensively studied effects in the word recognition literature. Using both conventional analyses of means and converging analyses (ex-Gaussian, quantile, diffusion modeling) of RT distributions, the present study provides a clearer understanding of the intriguing dissociation between semantic priming and different variables that increase the difficulty of the lexical decision task. Specifically, semantic priming produces interactive effects with stimulus quality (primarily in the tail of the RT distribution) and additive effects with nonword difficulty. In contrast, word-frequency produces the opposite pattern of joint effects, that is, word-frequency produces additive effects with stimulus quality, and interacts with nonword difficulty, again primarily in the tail of the distribution (see Yap et al., 2006). We are led to a descriptive framework that relies on distinct and interactive stages of processing leading from input to decision in the lexical decision task. Of course, the present account is based on a metaphorical description of the stages and processes that we and others (e.g., Borowsky & Besner, 1993; McNamara, 2005; Stolz & Besner, 1998) believe are useful for explaining the complex set of results that have now been well-established in the rich visual word recognition literature. We have not endorsed a particular computational approach in accommodating the present set of results, because it is likely that while extant models may capture a subset of findings, other findings may fall outside the scope of a given model. Indeed, these latter findings may reflect specific stages of processing brought on-line by the demands of the lexical decision task. We look forward to further explorations of these intriguing results within extant models.

References Anderson, J. R., & Milson, R. (1989). Human memory: An adaptive perspective. Psychological Review, 96, 703–719. doi:10.1037/0033295X.96.4.703 Ashby, F. G. (1982). Deriving exact predictions from the cascade model. Psychological Review, 89, 599 – 607. doi:10.1037/0033-295X.89.5.599 Balota, D. A., & Abrams, R. A. (1995). Mental chronometry: Beyond onset latencies in the lexical decision task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 1289 –1302. doi:10.1037/ 0278-7393.21.5.1289 Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage. Journal of Experimental Psychology: Human Perception and Performance, 10, 340 –357. doi:10.1037/0096-1523.10.3.340 Balota, D. A., & Spieler, D. H. (1999). Word frequency, repetition, and lexicality effects in word recognition tasks: Beyond measures of central tendency. Journal of Experimental Psychology: General, 128, 32–55. doi:10.1037/0096-3445.128.1.32 Balota, D. A., & Yap, M. J. (2006). Attentional control and flexible lexical processing: Explorations of the magic moment of word recognition. In S. Andrews (Ed.), From inkmarks to ideas: Current issues in lexical processing (pp. 229 –258). Hove, England: Psychology Press. Balota, D. A., & Yap, M. J. (2011). Moving beyond the mean in studies of mental chronometry: The power of response time distributional analyses.

14


Current Directions in Psychological Science, 20, 160 –166. doi:10.1177/ 0963721411408885 Balota, D. A., Yap, M. J., Cortese, M. J., Hutchison, K. A., Kessler, B., Loftis, B., . . . Treiman, R. (2007). The English Lexicon Project. Behavior Research Methods, 39, 445– 459. doi:10.3758/BF03193014 Balota, D. A., Yap, M. J., Cortese, M. J., & Watson, J. M. (2008). Beyond mean response latency: Response time distributional analyses of semantic priming. Journal of Memory and Language, 59, 495–523. doi: 10.1016/j.jml.2007.10.004 Becker, C. A. (1979). Semantic context and word frequency effects in visual word recognition. Journal of Experimental Psychology: Human Perception and Performance, 5, 252–259. doi:10.1037/0096-1523 .5.2.252 Becker, C. A., & Killion, T. H. (1977). Interaction of visual and cognitive effects in word recognition. Journal of Experimental Psychology: Human Perception and Performance, 3, 389 – 401. doi:10.1037/00961523.3.3.389 Besner, D. (1983). Basic decoding components in reading: Two dissociable feature extraction processes. Canadian Journal of Psychology/Revue canadienne de psychologie, 37, 429 – 438. doi:10.1037/h0080739 Besner, D., & Borowsky, R. (2006). Postscript: Plaut and Booth’s (2006) new simulations: What have we learned? Psychological Review, 113, 194 –195. doi:10.1037/0033-295X.113.1.194 Besner, D., & Smith, M. C. (1992). Models of visual word recognition: When obscuring the stimulus yields a clearer view. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 468 – 482. doi:10.1037/0278-7393.18.3.468 Besner, D., & Swan, M. (1982). Models of lexical access in visual word recognition. The Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 34(A), 313–325. Besner, D., Wartak, S., & Robidoux, S. (2008). Constraints on implemented models of basic processes in reading. Journal of Experimental Psychology: Human Perception and Performance, 34, 242–250. Bodner, G. E., & Masson, M. E. J. (1997). Masked repetition priming of words and nonwords: Evidence for a nonlexical basis for priming. Journal of Memory and Language, 37, 268 –293. doi:10.1006/ jmla.1996.2507 Bodner, G. E., Masson, M. E. J., & Richard, N. T. (2006). Repetition proportion biases masked priming of lexical decisions. Memory & Cognition, 34, 1298 –1311. doi:10.3758/BF03193273 Borowsky, R., & Besner, D. (1993). Visual word recognition: A multistage activation model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 813– 840. doi:10.1037/0278-7393.19.4.813 Borowsky, R., & Besner, D. (2006). Parallel distributed processing and lexical-semantic effects in visual word recognition: Are a few stages necessary? Psychological Review, 113, 181–193. doi:10.1037/0033-295X .113.1.181 Brown, M., & Besner, D. (2002). Semantic priming: On the role of awareness in visual word recognition in the absence of an expectancy. Consciousness and Cognition, 11, 402– 422. doi:10.1016/S10538100(02)00008-9 Brown, M., Stolz, J., & Besner, D. (2006). Dissociative effects of stimulus quality on semantic and morphological contexts in visual word recognition. Canadian Journal of Experimental Psychology/Revue canadienne de psychologie expe´rimentale, 60, 190 –199. doi:10.1037/ cjep2006018 Coane, J. H., & Balota, D. A. (2011). Face (and nose) priming for book: The malleability of semantic memory. Experimental Psychology, 58, 62–70. doi:10.1027/1618-3169/a000068 Cortese, M. J., & Fugett, A. (2004). Imageability ratings for 3,000 monosyllabic words. Behavior Research Methods, Instruments, & Computers, 36, 384 –387. doi:10.3758/BF03195585 Cortese, M. J., & Khanna, M. M. (2007). Age of acquisition predicts naming and lexical decision performance above and beyond 22 other

predictor variables: An analysis of 2,342 words. The Quarterly Journal of Experimental Psychology, 60, 1072–1082. doi:10.1080/ 17470210701315467 Cousineau, D., Brown, S. D., & Heathcote, A. (2004). Fitting distributions using maximum likelihood: Methods and packages. Behavior Research Methods, Instruments, & Computers, 36, 742–756. doi:10.3758/ BF03206555 den Heyer, K., & Goring, A. (1985). Semantic priming and word repetition: The two effects are additive. Journal of Memory and Language, 24, 699 –716. doi:10.1016/0749-596X(85)90054-3 Dufau, S., Grainger, J., & Ziegler, J. C. (2012). How to say “no” to a nonword: A leaky competing accumulator model of lexical decision. Journal of Experimental Psychology: Learning, Memory, and Cognition. Advance online publication. doi:10.1037/a0026948 Dun˜abeitia, J. A., Avile´s, A., & Carreiras, M. (2008). NoA’s ark: Influence of the number of associates in visual word recognition. Psychonomic Bulletin & Review, 15, 1072–1077. doi:10.3758/PBR.15.6.1072 Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 680 – 698. doi:10.1037/02787393.10.4.680 Forster, K. I., Davis, C., Schoknecht, C., & Carter, R. (1987). Masked priming with graphemically related forms: Repetition or partial activation? The Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 39(A), 211–251. doi:10.1080/ 14640748708401785 Gomez, P. (in press). Mathematical models of the lexical decision task. In J. S. Adelman (Ed.), Visual word recognition: Models and methods, orthography and phonology (Vol. 1). Florence, KY: Psychology Press. Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual word recognition: A multiple read-out model. Psychological Review, 103, 518 –565. doi:10.1037/0033-295X.103.3.518 Heathcote, A., & Brown, S. (2004). Reply to Speckman and Rouder: A theoretical basis for QML. Psychonomic Bulletin & Review, 11, 577– 578. doi:10.3758/BF03196614 Heathcote, A., Brown, S., & Mewhort, D. J. K. (2002). Quantile maximum likelihood estimation of response time distributions. Psychonomic Bulletin & Review, 9, 394 – 401. doi:10.3758/BF03196299 Hutchison, K. A. (2003). Is semantic priming due to association strength or feature overlap? A microanalytic review. Psychonomic Bulletin & Review, 10, 785– 813. doi:10.3758/BF03196544 Jacobs, A. M., & Grainger, J. (1994). Models of visual word recognition: Sampling the state of the art. Journal of Experimental Psychology: Human Perception and Performance, 20, 1311–1334. doi:10.1037/ 0096-1523.20.6.1311 James, C. T. (1975). The role of semantic information in lexical decisions. Journal of Experimental Psychology: Human Perception and Performance, 104, 130 –136. Joordens, S., & Becker, S. (1997). The long and short of semantic priming effects in lexical decision. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 1083–1105. doi:10.1037/02787393.23.5.1083 Juhasz, B. J. (2005). Age-of-acquisition effects in word and picture identification. Psychological Bulletin, 131, 684 –712. doi:10.1037/00332909.131.5.684 Juhasz, B. J., Yap, M. J., Dicke, J., Taylor, S. C., & Gullick, M. M. (2011). Tangible words are recognized faster: The grounding of meaning in sensory and perceptual systems. The Quarterly Journal of Experimental Psychology, 64, 1683–1691. doi:10.1080/17470218.2011.605150 Lund, K., & Burgess, C. (1996). Producing high-dimensional semantic spaces from lexical co-occurrence. Behavior Research Methods, Instruments, & Computers, 28, 203–208. doi:10.3758/BF03204766 Lupker, S. J., & Pexman, P. M. (2010). Making things difficult in lexical decision: The impact of pseudohomophones and transposed-letter non-

ADDITIVE AND INTERACTIVE EFFECTS words on frequency and semantic priming effects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36, 1267–1289. doi:10.1037/a0020125 Masson, M. E. J. (1995). A distributed memory model of semantic priming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 3–23. doi:10.1037/0278-7393.21.1.3 Matzke, D., & Wagenmakers, E.-J. (2009). Psychological interpretation of the ex-Gaussian and shifted Wald parameters: A diffusion model analysis. Psychonomic Bulletin & Review, 16, 798 – 817. doi:10.3758/ PBR.16.5.798 McClelland, J. L. (1979). On the time relations of mental processes: An examination of systems of processes in cascade. Psychological Review, 86, 287–330. doi:10.1037/0033-295X.86.4.287 McNamara, T. P. (2005). Semantic priming: Perspectives from memory and word recognition. doi:10.4324/9780203338001 Meyer, D. E., Schvaneveldt, R. W., & Ruddy, M. G. (1975). Loci of contextual effects on visual word-recognition. In P. M. A. Rabbitt (Ed.), Attention and performance V (pp. 98 –118). London, England: Academic Press. Milota, V. C., Widau, A. A., McMickell, M. R., Juola, J. F., & Simpson, G. B. (1997). Strategic reliance on phonological mediation in lexical access. Memory & Cognition, 25, 333–344. doi:10.3758/BF03211289 Morton, J. (1969). The interaction of information in word recognition. Psychological Review, 76, 165–178. doi:10.1037/h0027366 Murray, W. S., & Forster, K. I. (2004). Serial mechanisms in lexical access: The rank hypothesis. Psychological Review, 111, 721–756. doi: 10.1037/0033-295X.111.3.721 Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. Besner & G. W. Humphreys (Eds.), Basic processes in reading (pp. 264 –336). Hillsdale, NJ: Erlbaum. Neely, J. H., & Keefe, D. E. (1989). Semantic context effects on visual word processing: A hybrid prospective/retrospective processing theory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 24, pp. 207–248). New York, NY: Academic Press. Neely, J. H., Keefe, D. E., & Ross, K. L. (1989). Semantic priming in the lexical decision task: Roles of prospective prime-generated expectancies and retrospective semantic matching. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 1003–1019. doi:10.1037/ 0278-7393.15.6.1003 Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (2004). The University of South Florida word association, rhyme, and word fragment norms. Behavior Research Methods, Instruments, & Computers, 36, 402– 407. doi:10.3758/BF03195588 Norris, D. (2006). The Bayesian reader: Explaining word recognition as an optimal Bayesian decision process. Psychological Review, 113, 327– 357. doi:10.1037/0033-295X.113.2.327 Norris, D. (2009). Putting it all together: A unified account of word recognition and reaction-time distributions. Psychological Review, 116, 207–219. doi:10.1037/a0014259 O’Malley, S., Reynolds, M., & Besner, D. (2007). Qualitative differences between the joint effects of stimulus quality and word frequency in lexical decision and reading aloud: Extensions to Yap and Balota. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 451– 458. doi:10.1037/0278-7393.33.2.451 Perea, M., & Lupker, S. J. (2003). Transposed-letter confusability effects in masked form priming. In S. Kinoshita & S. J. Lupker (Eds.), Masked priming: State of the art (pp. 97–120). Hove, England: Psychology Press. Perea, M., & Lupker, S. J. (2004). Can CANISO activate CASINO? Transposed-letter similarity effects with nonadjacent letter positions. Journal of Memory and Language, 51, 231–246. doi:10.1016/ j.jml.2004.05.005 Perfetti, C. A., & Hart, L. (2002). The lexical quality hypothesis. In L.

15

Vehoeven, C. Elbro, & P. Reitsma (Eds.), Precursors of functional literacy (pp. 189 –213). Amsterdam, the Netherlands: John Benjamins. Pexman, P. M., & Lupker, S. J. (1999). The impact of semantic ambiguity on visual word recognition: Do homophone and polysemy effects cooccur? Canadian Journal of Experimental Psychology/Revue canadienne de psychologie expe´rimentale, 53, 323–334. doi:10.1037/ h0087320 Pexman, P. M., Lupker, S. J., & Hino, Y. (2002). The impact of feedback semantics in visual word recognition: Number of features effects in lexical decision and naming tasks. Psychonomic Bulletin & Review, 9, 542–549. doi:10.3758/BF03196311 Plaut, D. C. (1997). Structure and function in the lexical system: Insights from distributed models of word reading and lexical decision. Language and Cognitive Processes, 12, 765– 806. doi:10.1080/016909697386682 Plaut, D. C., & Booth, J. R. (2000). Individual and developmental differences in semantic priming: Empirical and computational support for a single-mechanism account of lexical processing. Psychological Review, 107, 786 – 823. doi:10.1037/0033-295X.107.4.786 Plaut, D. C., & Booth, J. R. (2006). More modeling but still no stages: Reply to Borowsky and Besner. Psychological Review, 113, 196 –200. doi:10.1037/0033-295X.113.1.196 Plourde, C. E., & Besner, D. (1997). On the locus of the word frequency effect in visual word recognition. Canadian Journal of Experimental Psychology/Revue canadienne de psychologie expe´rimentale, 51, 181– 194. doi:10.1037/1196-1961.51.3.181 Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85, 59 –108. doi:10.1037/0033-295X.85.2.59 Ratcliff, R. (1979). Group reaction time distributions and an analysis of distribution statistics. Psychological Bulletin, 86, 446 – 461. doi: 10.1037/0033-2909.86.3.446 Ratcliff, R., Gomez, P., & McKoon, G. (2004). A diffusion model account of the lexical decision task. Psychological Review, 111, 159 –182. doi: 10.1037/0033-295X.111.1.159 Ratcliff, R., & McKoon, G. (1988). A retrieval theory of priming in memory. Psychological Review, 95, 385– 408. doi:10.1037/0033295X.95.3.385 Ratcliff, R., Thapar, A., & McKoon, G. (2010). Individual differences, aging, and IQ in two-choice tasks. Cognitive Psychology, 60, 127–157. doi:10.1016/j.cogpsych.2009.09.001 Roberts, S., & Sternberg, S. (1993). The meaning of additive reaction-time effects: Tests of three alternatives. In D. E. Meyer & S. Kornblum (Eds.), Attention and Performance XIV: Synergies in experimental psychology, artificial intelligence, and cognitive neuroscience—A Silver Jubilee (pp. 611– 653). Cambridge, MA: MIT Press. Schmiedek, F., Oberauer, K., Wilhelm, O., Su¨␤, H.-M., & Wittmann, W. W. (2007). Individual differences in components of reaction time distributions and their relations to working memory and intelligence. Journal of Experimental Psychology: General, 136, 414 – 429. doi: 10.1037/0096-3445.136.3.414 Schneider, W., Eschman, A., & Zuccolotto, A. (2001). E-prime user’s guide. Pittsburgh, PA: Psychology Software Tools. Shipley, W. C. (1940). A self-administering scale for measuring intellectual impairment and deterioration. Journal of Psychology: Interdisciplinary and Applied, 9, 371–377. doi:10.1080/00223980.1940.9917704 Shulman, H. G., & Davison, T. C. (1977). Control properties of semantic coding in a lexical decision task. Journal of Verbal Learning and Verbal Behavior, 16, 91–98. doi:10.1016/S0022-5371(77)80010-8 Siakaluk, P. D., Pexman, P. M., Aguilera, L., Owen, W. J., & Sears, C. R. (2008). Evidence for the activation of sensorimotor information during visual word recognition: The body-object interaction effect. Cognition, 106, 433– 443. doi:10.1016/j.cognition.2006.12.011 Speckman, P. L., & Rouder, J. N. (2004). A comment on Heathcote, Brown, and Mewhort’s QMLE method for response time distributions.

16


Psychonomic Bulletin & Review, 11, 574 –576. doi:10.3758/ BF03196613 Spieler, D. H., Balota, D. A., & Faust, M. E. (2000). Levels of selective attention revealed through analyses of response time distributions. Journal of Experimental Psychology: Human Perception and Performance, 26, 506 –526. doi:10.1037/0096-1523.26.2.506 Stanners, R. F., Jastrzembski, J. E., & Westbrook, A. (1975). Frequency and visual quality in a word-nonword classification task. Journal of Verbal Learning and Verbal Behavior, 14, 259 –264. doi:10.1016/ S0022-5371(75)80069-7 Staub, A. (2011). The effect of lexical predictability on distributions of eye fixation durations. Psychonomic Bulletin & Review, 18, 371–376. doi: 10.3758/s13423-010-0046-9 Sternberg, S. (1969). The discovery of processing stages: Extensions of Donders’ method. Acta Psychologica, 30, 276 –315. doi:10.1016/00016918(69)90055-9 Sternberg, S. (1998). Discovering mental processing stages: The method of additive factors. In D. Scarborough & S. Sternberg (Eds.), Methods, models, and conceptual issues: An invitation to cognitive science (pp. 703– 863). Cambridge, MA: MIT Press. Stolz, J. A., & Besner, D. (1998). Role of set in visual word recognition: Activation and activation blocking as nonautomatic processes. Journal of Experimental Psychology: Human Perception and Performance, 24, 1642–1655. doi:10.1037/0096-1523.24.6.1642 Stolz, J. A., & Neely, J. H. (1995). When target degradation does and does not enhance semantic context effects in word recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 596 – 611. doi:10.1037/0278-7393.21.3.596 Stone, G. O., & Van Orden, G. C. (1992). Resolving empirical inconsistencies concerning priming, frequency, and nonword foils in lexical decision. Language and Speech, 35, 295–324. Stone, G. O., & Van Orden, G. C. (1993). Strategic control of processing in word recognition. Journal of Experimental Psychology: Human Perception and Performance, 19, 744 –774. doi:10.1037/0096-1523 .19.4.744 Tenpenny, P. L. (1995). Abstractionist versus episodic theories of repetition priming and word identification. Psychonomic Bulletin & Review, 2, 339 –363. doi:10.3758/BF03210972 Thomas, M. A., Neely, J. H., & O’Connor, P. (2012). When word identification gets tough, retrospective semantic processing comes to the rescue. Journal of Memory and Language. Advance online publication. doi:10.1016/j.jml.2012.02.002 Versace, R., & Nevers, B. (2003). Word frequency effect on repetition

priming as a function of prime duration and delay between the prime and the target. British Journal of Psychology, 94, 389 – 408. doi:10.1348/ 000712603767876299 Voss, A., & Voss, J. (2007). Fast-dm: A free program for efficient diffusion model analysis. Behavior Research Methods, 39, 767–775. doi: 10.3758/BF03192967 Voss, A., & Voss, J. (2008). A fast numerical algorithm for the estimation of diffusion model parameters. Journal of Mathematical Psychology, 52, 1–9. doi:10.1016/j.jmp.2007.09.005 White, S. J., & Staub, A. (2011). The distribution of fixation durations during reading: Effects of stimulus quality. Journal of Experimental Psychology: Human Perception and Performance. Advance online publication. doi:10.1037/a0025338 Whittlesea, B. W. A., & Jacoby, L. (1990). Interaction of prime repetition with visual degradation: Is priming a retrieval phenomenon? Journal of Memory and Language, 29, 546 –565. doi:10.1016/0749-596X(90) 90051-Z Yap, M. J., & Balota, D. A. (2007). Additive and interactive effects on response time distributions in visual word recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 274 – 296. doi:10.1037/0278-7393.33.2.274 Yap, M. J., & Balota, D. A. (2009). Visual word recognition of multisyllabic words. Journal of Memory & Language, 60, 502–529. Yap, M. J., Balota, D. A., Cortese, M. J., & Watson, J. M. (2006). Single versus dual process models of lexical decision performance: Insights from RT distributional analysis. Journal of Experimental Psychology: Human Perception and Performance, 32, 1324 –1344. doi:10.1037/ 0096-1523.32.6.1324 Yap, M. J., Balota, D. A., Tse, C.-S., & Besner, D. (2008). On the additive effects of stimulus quality and word frequency in lexical decision: Evidence for opposing interactive influences revealed by RT distributional analyses. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34, 495–513. doi:10.1037/0278-7393.34.3.495 Yap, M. J., Tse, C.-S., & Balota, D. A. (2009). Individual differences in the joint effects of semantic priming and word frequency: The role of lexical integrity. Journal of Memory and Language, 61, 303–325. doi:10.1016/ j.jml.2009.07.001 Yarkoni, T., Balota, D. A., & Yap, M. J. (2008). Beyond Coltheart’s N: A new measure of orthographic similarity. Psychonomic Bulletin & Review, 15, 971–979. doi:10.3758/PBR.15.5.971 Zachary, R. A. (1992). Shipley Institute of Living Scale—Revised manual. Los Angeles, CA: Western Psychological Services.

(Appendix follows)


17

Appendix Stimuli in Experiments 1 and 2 PRIME

TARGET

PRIME

TARGET

PRIME

TARGET

MINDED MISTREAT SUGGESTION PHYSIOLOGY RENOUNCE AGGRAVATE BLUEPRINT DESCEND AMAZE TUTU ROUGE COMPUTE DIAGRAM GOURMET MONKEY CHOIR VACUUM CASHIER MIX FUSS UNDERSTAND LINK ARTS IMAGINATION KNIT BETRAY INCREASE VICTORY THAW REPUBLICAN VARY INVENT DISGRACE HONEST LIKE DISINTEGRATE SWIMMER MULTIPLY FLIPPER MULE

ABSENT ABUSE ADVISE ANATOMY ANNOUNCE ANNOY ARCHITECT ASCEND AWE BALLET BLUSH CALCULATE CHART CHEF CHIMPANZEE CHORUS CLEANER CLERK COMBINE COMPLAIN COMPREHEND CONNECT CRAFTS CREATIVITY CROCHET DECEIVE DECREASE DEFEAT DEFROST DEMOCRAT DIFFER DISCOVER DISGUST DISHONEST DISLIKE DISSOLVE DIVER DIVIDE DOLPHIN DONKEY

BIT HUMILIATE HUG JEALOUSY DIMINISH PHYSICAL PASTE GANDER PASTURE CHARCOAL NAIL CROPS HITCH CLUE HOSTESS NOTIFY WOODWIND OFFEND CREATE RUN POUCH OZONE SERMON CHAIN FLUID FORTUNE BUTLER TUPPERWARE PLUS DIVIDE ARTIST PINK GUARDIAN HESITATE ALLOW WORD ELECTRICIAN OUNCE RECOGNITION COPIER

DRILL EMBARRASS EMBRACE ENVY FADE FITNESS GLUE GOOSE GRAZE GRILL HAMMER HARVEST HIKE HINT HOST INFORM INSTRUMENT INSULT INVENT JOG KANGAROO LAYER LECTURE LINK LIQUID LOTTERY MAID MICROWAVE MINUS MULTIPLY PAINTER PANTHER PARENT PAUSE PERMIT PHRASE PLUMBER POUND PRAISE PRINTER

HALLOWEEN VIOLET FURY ACCEPT BEEF DECAY SPOILED SELLER UPSTREAM SUB REPRIMAND EGGS WALRUS GROW DRAW SLOPE TECHNIQUE SLENDER SUFFOCATE DOZE MASSAGE REAP ROT DETECTIVE POSTAGE PILOT ADD SCALPEL BROOM LIGHTNING NEAT FLOOR UNDERGROUND SWEEP HERO SERVER WAITER BLUBBER EVIL LENGTH

PUMPKIN PURPLE RAGE REJECT ROAST ROT ROTTEN SALESMAN SALMON SANDWICH SCOLD SCRAMBLE SEAL SHRINK SKETCH SKI SKILL SLIM SMOTHER SNOOZE SOOTHE SOW SPOIL SPY STAMP STEWARDESS SUBTRACT SURGEON SWEEP THUNDER TIDY TILE TUNNEL VACUUM VILLAIN WAITER WAITRESS WHALE WICKED WIDTH

(Appendix continues)


18 Appendix (continued) PRIME

LEGAL NW

PRIME

LEGAL NW

PRIME

LEGAL NW

SCALDED MERELY SATISFIED ADHESIVE EXERCISE REFRAIN POLLS PIPE SNIP MOBSTER HORRORS GENTLE CLINIC RESENTFUL NUT INDIVIDUAL SHOWN MELLOW FORK MYSTERIOUS CHEAPER DEEPER FESTIVAL COUNTRIES FLASHBACK SPRINKLE CRUSHED TONE TECHNOLOGY TILTED THEREIN ALTO HANDING EARN GOSSIPING REDUCE TRANSPORT RUB OPTIMISTIC BANQUET

ABBLAIMS ABLOW ABTRAY AIG ALVEOBA ASTITE ATTOND ATTOW BAGGOT BANDOD BEEHOVE BLIFFER BLOWNINGLY BLUPID BROWERING BRUNK CARNAY CERSERK CHIGS CHIRM CHRUTTER CHUGGLE CLANSFER CLEB CLERMOS CLIRMISHES DALTOP DEPATE DETROTH DILT DORST DRALLER DRAMENS DRATTY DRELL DROCKER ESIN FERBS FIGHTEN FLINDLE

ANGER PIN PRIVACY REEF PUNK VESSEL TOUGHEN GLORIOUS DETOUR CURE BRIDE ZERO POODLE CREWS BILLION TOY DIM LAD AMBITION INVENTORY CIVILIZED SQUANDER IDLE TIMING JIGSAW MOTION LINING FROZE INVITATION ALIENS SUSPICIOUS QUOTA ANGLE CHAT SPADE CLUSTER FOREHEAD POSED SOAP LIDS

FOLT FRUNKER GAISE GIRES GLERVING GLIP GRASH HANDID HIFE HOAL HUPS INTRALECTS IRRITERACY JOP LANDIT LARGIN LER LICKET LICYCLE LIDEN LINTER MENCHES MIFER MOIL MOT MURDINESS MURGLARY NALLOW NOTARUSE OPPASION PACKLES PAULT PHRAWL PLARTLY PLAUGABLE PLIGOTS PLINDED PLURIOUS PRAKES PRANKFUL

DATA EXPANSION RISKY LUNGS ZEBRA PANCAKE REMARKABLE SCALP FOLDING AMPLE COMEDY SHADOW DECK HELICOPTER LABELS PORTION PRYING GOAL DWELLING TALENTS SPENDS STEW KICKING QUANTITY TANKER GARBLED INCENSE COMMERCIAL SINGLED CEREAL LION DRIVEWAY OILY ROB SOBS MASS SCARF FLOURISH SONS WALTZ

PRINGS PROWN PRUEAL PUDGET RACUST ROG SCHINTER SCOVENANCE SCRASH SETRAYS SHUMP SILD SLONK SLOUND SLOUP SMASTOR SMICKER SNI SNILL SONE SPOVE STECIPICE STORIFIED SUSHER SUTTRESS TACE TAULDRON THILING THRIP TRAVELAN TRINACH TROKER TROOGE TROST TUAL TULL UNLUND VORGAL WALPS WOCKED

(Appendix continues)


19

Appendix (continued) PRIME

TL NW

PRIME

TL NW

PRIME

TL NW

SQUANDER INVENTORY TALENTS SCALDED SPADE NUT CLUSTER REMARKABLE HORRORS DRIVEWAY ALTO GENTLE OILY SPENDS CIVILIZED MASS SOBS PORTION SCALP DWELLING SONS QUANTITY PRIVACY AMPLE OPTIMISTIC JIGSAW ANGER SINGLED LION VESSEL BRIDE FROZE ANGLE SHOWN TILTED GOSSIPING SPRINKLE SCARF TOY WALTZ

ABN ADMINSITER ADVSIOR ALD AMSUE AUMSE BAGDE BAKSET BANR BASBEALL BEAHLF BLAEK BLESSNIG BLIDNED BOBMS BOUELVARD BREAHTER BUGIGNG CASAUL CATROON CELLOHPANE CHAIROT CHEIMST CHEKES CHIARS CILP CLIAMTE CONENCT COPMASS COUSNEL CRITREIA CRON CUHCK DAERD DEABTE DEDALY DEINAL DETETNION DETREMINE DISCILPINE

AMBITION CHEAPER KICKING CREWS LIDS LAD RISKY COMMERCIAL POSED FLASHBACK RUB SHADOW GLORIOUS FOLDING SNIP SATISFIED COMEDY THEREIN GARBLED HELICOPTER FLOURISH MERELY EXPANSION LUNGS ADHESIVE INCENSE REDUCE CRUSHED SOAP CHAT MYSTERIOUS BILLION TECHNOLOGY LABELS POLLS REFRAIN EXERCISE RESENTFUL PIN HANDING

DISCNONECT DUTSY EANREST ELETCRIC ELPOE ETIQEUTTE EXPNAD FAIERST FECTH FLWOER FOERST FOOELD FRAMGENT GIGNER GIRP GMY GRUGDE HOKCEY HOTSILE HUDRLE INGREDEINT IRM JEKRED LANUCH LAOF LEAHTER LOACTE LODULY LOGDE LSATS LYIRC MAITNAIN MAKRED MASSGAE MCOK MNEU MZAE OWEND PAESANT PEIRSH

ALIENS CEREAL REEF FORK CURE IDLE PRYING GOAL DATA SUSPICIOUS MOBSTER ZEBRA DIM FOREHEAD LINING COUNTRIES DETOUR PUNK STEW POODLE MELLOW MOTION BANQUET QUOTA ZERO PIPE INDIVIDUAL FESTIVAL TIMING CLINIC PANCAKE ROB TANKER TRANSPORT TOUGHEN INVITATION DECK TONE DEEPER EARN

PORM POVRETY PULBISH QEUER QUATNITY REA RECEPIT RESREVE RIBOBN RUOMR SAOP SAPMLE SARACSTIC SATLL SCAK SHERWD SHOLEACE SLUBMER SPATSIC SPLEDNID SPRAK STIE STRATLE SUACE SUCECEDS SUOR SVAES SWRON SYL TAEKR TEPMER THRETAEN TIDNIGS TRGAIC VANQIUSH VIAN VITRUE WAETRS WIERD WOVLES

Note. NW ⫽ nonword; TL ⫽ transposed-letter.

Received September 26, 2011 Revision received March 22, 2012 Accepted April 2, 2012 䡲